def color_map_by_land_water(spm_map, scene_path):
# color code image by cloud and land
all_imgs = lgd.get_scene_imgs(scene_path)
# read cloud image
cloud_path = lgd.get_cloud(all_imgs)
cloud = cv2.imread(cloud_path, cv2.IMREAD_ANYDEPTH)
cloud = np.array((cloud / 255.), dtype='uint8')
# make clouds white
spm_map[cloud == 1] = 1.0
# read water image
water = cv2.imread(
'/Users/Nathan/Dropbox/SedimentLearning/data/landsat/LT50440342010098-SC20160218112217/LT50440342010098PAC01_sr_land_water_qa.tif',
cv2.IMREAD_ANYCOLOR)
water = np.array((water / 255.), dtype='uint8')
# make land mass black
spm_map[water == 0] = 0
# make land and cloud gray
spm_map[np.multiply((1 - water), cloud) == 1] = 0.8
return spm_map
def create_color_map(scene_path=''):
'''
Uses the rgb bands to reconstruct a 'true color' satellite image
:param scene_path:
:return:
'''
scene_data, image_shape, valid = get_feature_array_L57(scene_path)
color_img = np.zeros((image_shape[0], image_shape[1], 3))
color_img[:, :, 0] = scene_data[:, 0].reshape(image_shape) # blue - band 1
color_img[:, :, 1] = scene_data[:, 1].reshape(image_shape) # green - band 2
color_img[:, :, 2] = scene_data[:, 2].reshape(image_shape) # red - band 3
# scale values
color_img = color_img * 255. / 4000.
all_imgs = lgd.get_scene_imgs(scene_path)
scene_name = lgd.get_scene_name(all_imgs[0])
cv2.imwrite('../figures/color_map_{}.jpg'.format(scene_name), color_img)
def create_spm_map(theta=None, scene_path='', log_spm_flag=False, color_flag=True):
# TODO add actual spm values to legend
'''
For info on sr_cloud_qa and sr_land_water_qa image values
http://landsat.usgs.gov/landsat_climate_data_records_quality_calibration.php
cloud: 255 = cloud, water: 255 = water
:param theta: theta for the linear model
:param scene_path: full path to the scene folder
:param log_spm_flag: flag for whether to map spm or log(spm), true means log(spm)
:return:
'''
if (theta is None):
theta = create_model()
else:
print('Using given theta. Creating predicted SPM map')
scene_data, image_shape, valid = get_feature_array_L57(scene_path)
predicted_spm_log = np.dot(scene_data, theta)
predicted_spm = np.exp(predicted_spm_log)
# map spm map to -1 to 1 range
if (log_spm_flag):
spm_map = predicted_spm_log.reshape(image_shape)
# high log(spm) values are around 4 so map (-1)-4 to -1-1
spm_map = np.interp(spm_map, (-1, 4), (-1, 1))
else:
spm_map = predicted_spm.reshape(image_shape)
# high spm values are around 50
# map 0-50 to -1 to 1
spm_map = np.interp(spm_map, (0, 50), (-1, 1))
print('Done creating predicted SPM map')
# map spm to jet color scale, exclude the alpha channel
spm_map = plt.cm.jet(spm_map)[:, :, 0:3]
# color code the map: make land black, make cloud white
spm_map = color_map_by_land_water(spm_map, scene_path)
print 'Done color coding map'
# create figure
dpi = 400
plt.figure(dpi=dpi)
plt.imshow(spm_map, interpolation='nearest')
c = plt.colorbar(ticks=[0, 1.0])
# add labels depending on log or not
if (log_spm_flag):
c.set_ticklabels(['ln(SPM) < -1', 'ln(SPM) > 4'])
else:
c.set_ticklabels(['SPM < 0', 'SPM > 50'])
# figure out the name
all_imgs = lgd.get_scene_imgs(scene_path)
scene_name = lgd.get_scene_name(all_imgs[0])
log_str = ('log_' if log_spm_flag else '')
folder = log_str + 'colormap_spm'
datetime = lgd.get_datetime_from_metadata(lgd.get_metadata_path(scene_path))
# Add title
plt.title('Map of SPM in SF Bay: ' + str(scene_name) + '\n' + str(datetime) + ' GMT')
# save figure
plt.savefig('../figures/{}/{}{}.jpg'.format(folder, log_str, scene_name), dpi=dpi)
print 'Done saving map\n'
def get_feature_array_L57(scene_folder_path):
# TODO: fill this out
'''
ONLY WORKS FOR LANDSAT 5 and 7 images. Landsat 8 images have different band wavelengths
Load the images in the scene folder into an array that the model can read. The format is:
['reflec_1', 'reflec_2', 'reflec_3', 'reflec_4', 'reflec_5', 'reflec_7', ratio1, ratio2, ratio3, ratio4, ratio5]
:param scene_folder_path: full path to folder of scene
:return:
'''
print 'Getting features from scene images'
# data to load
data_columns = ['reflec_1', 'reflec_2', 'reflec_3', 'reflec_4', 'reflec_5', 'reflec_7']
data = {}
for key in data_columns:
data[key] = np.array([])
image_shape = np.nan
# get names of all images in the scene folder
imgs = lgd.get_scene_imgs(scene_folder_path)
# get the paths for the different types of image
sr_paths = lgd.get_sr_band_imgs(imgs)
for path, band in zip(sr_paths, data_columns):
# print path, band
data[band] = cv2.imread(path, cv2.IMREAD_ANYDEPTH)
if image_shape is np.nan:
image_shape = data[band].shape
# print image_shape
data[band] = data[band].ravel()
full_data = np.zeros((6, data['reflec_1'].size))
# put data in correct order
full_data[0] = data['reflec_1']
full_data[1] = data['reflec_2']
full_data[2] = data['reflec_3']
full_data[3] = data['reflec_4']
full_data[4] = data['reflec_5']
full_data[5] = data['reflec_7']
# Transpose the data
full_data = full_data.T
# make sure the first column is reflec 1
assert np.array_equal(full_data[:, 0], data['reflec_1'])
print('Loaded surface reflectance tifs. Calculating band ratios...')
full_data = regression.Kau_MB_BR_features(full_data)
# make sure the first column is still reflec 1
# print 'Sum square differences: ' + str(np.sum(np.power(full_data[:, 0] - data['reflec_1'], 2)))
assert np.array_equal(full_data[:, 0], data['reflec_1'])
print('Finished loading regression features: 3 band ratios + 4 surface reflectances')
# create matrix representing valid data
valid_data = np.ones_like(data['reflec_1'])
for band in range(4):
valid_data[full_data[:, band] < 0] = 0
valid_data[full_data[:, band] > 10000] = 0
return full_data, image_shape, valid_data
